From: "Zhang, Qi Z" <qi.z.zhang@intel.com>
To: "Cui, LunyuanX" <lunyuanx.cui@intel.com>,
"Yang, Qiming" <qiming.yang@intel.com>,
"Wu, Jingjing" <jingjing.wu@intel.com>,
"Xing, Beilei" <beilei.xing@intel.com>,
"Wang, ShougangX" <shougangx.wang@intel.com>,
"Stillwell Jr, Paul M" <paul.m.stillwell.jr@intel.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [RFC v3] net/i40e: enable multi-queue Rx interrupt for VF
Date: Tue, 17 Sep 2019 02:33:54 +0000 [thread overview]
Message-ID: <039ED4275CED7440929022BC67E7061153D97AEB@SHSMSX105.ccr.corp.intel.com> (raw)
In-Reply-To: <20190916144308.119187-1-lunyuanx.cui@intel.com>
> -----Original Message-----
> From: Cui, LunyuanX
> Sent: Monday, September 16, 2019 10:43 PM
> To: Yang, Qiming <qiming.yang@intel.com>; Wu, Jingjing
> <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>; Zhang, Qi Z
> <qi.z.zhang@intel.com>; Wang, ShougangX <shougangx.wang@intel.com>;
> Stillwell Jr, Paul M <paul.m.stillwell.jr@intel.com>
> Cc: dev@dpdk.org; Cui, LunyuanX <lunyuanx.cui@intel.com>
> Subject: [RFC v3] net/i40e: enable multi-queue Rx interrupt for VF
>
> Current implementation is that only one Rx queue can support interrupt,
> because all queues are mapped in the same vector id in vfio_enable_msix().
> So VF can not support multi-queue Rx interrupt in the interrupt mode.
>
> In this patch, if the packet I/O interrupt on datapath is enabled
> (rte_intr_dp_is_en(intr_handle) is true), we map different interrupt vectors to
> each queue and send this map to PF.
> After PF sets the map to the register,
> all Rx queue interrupts will be received.
>
> In addition, because of the i40e performance in ioctl(), the maximum
> supported interrupt vector id is 4.
> If vector id is more than 4, i40e driver will fail to start.
> So when queue number is more than 4,
> we set up a loop of interrupt vectors map from 1 to 4.
Don't know why we need to limit max interrupt to 4, why i40e driver will fail to start, could share more detail?
I think you can do the similar implementation as iavf driver does (ref iavf_config_irq_map)
>
> Signed-off-by: Lunyuan Cui <lunyuanx.cui@intel.com>
> ---
> v3:
> * combined 2 lines into 1 line
> before:
> map_info->vecmap[i].rxq_map = 0;
> map_info->vecmap[i].rxq_map |= 1 << i;
> after:
> map_info->vecmap[i].rxq_map = 1 << i;
>
> v2:
> * set up a loop of interrupt vectors map from 1 to 4, and sent
> message from VF to PF by one time.
> ---
> drivers/net/i40e/i40e_ethdev_vf.c | 27 +++++++++++++++++----------
> 1 file changed, 17 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/net/i40e/i40e_ethdev_vf.c
> b/drivers/net/i40e/i40e_ethdev_vf.c
> index 308fb9835..abb2d1353 100644
> --- a/drivers/net/i40e/i40e_ethdev_vf.c
> +++ b/drivers/net/i40e/i40e_ethdev_vf.c
> @@ -645,13 +645,15 @@ i40evf_configure_vsi_queues(struct rte_eth_dev
> *dev)
> return ret;
> }
>
> +#define RTE_LIBRTE_I40E_IRQ_NUM_PER_VF 4
> +
> static int
> i40evf_config_irq_map(struct rte_eth_dev *dev) {
> struct i40e_vf *vf =
> I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
> struct vf_cmd_info args;
> uint8_t cmd_buffer[sizeof(struct virtchnl_irq_map_info) + \
> - sizeof(struct virtchnl_vector_map)];
> + sizeof(struct virtchnl_vector_map) * dev->data->nb_rx_queues];
> struct virtchnl_irq_map_info *map_info;
> struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
> struct rte_intr_handle *intr_handle = &pci_dev->intr_handle; @@ -665,18
> +667,23 @@ i40evf_config_irq_map(struct rte_eth_dev *dev)
> vector_id = I40E_MISC_VEC_ID;
>
> map_info = (struct virtchnl_irq_map_info *)cmd_buffer;
> - map_info->num_vectors = 1;
> - map_info->vecmap[0].rxitr_idx = I40E_ITR_INDEX_DEFAULT;
> - map_info->vecmap[0].vsi_id = vf->vsi_res->vsi_id;
> - /* Alway use default dynamic MSIX interrupt */
> - map_info->vecmap[0].vector_id = vector_id;
> - /* Don't map any tx queue */
> - map_info->vecmap[0].txq_map = 0;
> - map_info->vecmap[0].rxq_map = 0;
> + map_info->num_vectors = dev->data->nb_rx_queues;
> for (i = 0; i < dev->data->nb_rx_queues; i++) {
> - map_info->vecmap[0].rxq_map |= 1 << i;
> + map_info->vecmap[i].rxitr_idx = I40E_ITR_INDEX_DEFAULT;
> + map_info->vecmap[i].vsi_id = vf->vsi_res->vsi_id;
> + /* Alway use default dynamic MSIX interrupt */
> + map_info->vecmap[i].vector_id = vector_id;
> + /* Don't map any tx queue */
> + map_info->vecmap[i].txq_map = 0;
> + map_info->vecmap[i].rxq_map = 1 << i;
> if (rte_intr_dp_is_en(intr_handle))
> intr_handle->intr_vec[i] = vector_id;
> + if (vector_id > I40E_MISC_VEC_ID) {
> + if (vector_id < RTE_LIBRTE_I40E_IRQ_NUM_PER_VF)
> + vector_id++;
> + else
> + vector_id = I40E_RX_VEC_START;
> + }
> }
>
> args.ops = VIRTCHNL_OP_CONFIG_IRQ_MAP;
> --
> 2.17.1
next prev parent reply other threads:[~2019-09-17 2:34 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-02 19:45 [dpdk-dev] [RFC] " lunyuan.cui
2019-09-16 14:43 ` [dpdk-dev] [RFC v3] " lunyuan.cui
2019-09-17 2:33 ` Zhang, Qi Z [this message]
2019-09-25 11:06 ` [dpdk-dev] [RFC v4] " lunyuan.cui
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=039ED4275CED7440929022BC67E7061153D97AEB@SHSMSX105.ccr.corp.intel.com \
--to=qi.z.zhang@intel.com \
--cc=beilei.xing@intel.com \
--cc=dev@dpdk.org \
--cc=jingjing.wu@intel.com \
--cc=lunyuanx.cui@intel.com \
--cc=paul.m.stillwell.jr@intel.com \
--cc=qiming.yang@intel.com \
--cc=shougangx.wang@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).